# Changelog All notable changes to napalm-device-types. The format follows [Keep a Changelog](https://keepachangelog.com/), versions follow [Semantic Versioning](https://semver.org/). Versions up to 4.21.0 were not recorded here; their history is the git log. ## [Unreleased] ### Added - `SshServerAuthMixin.get_ssh_server_auth()`: how an OpenSSH server lets people log in -- password, keyboard-interactive, public key, host-based and GSSAPI authentication, `AuthenticationMethods` as lists of methods, `PermitRootLogin` and `UsePAM`, from `sshd -T` as root, and sshd's PAM `auth` stack (each module with its control and arguments, includes followed), from `/etc/pam.d` as the login user. `OpenSshServerMixin` carries it, so every driver that can harden sshd has it. `parse_sshd_auth`, `parse_pam_auth`, `parse_pam_files`, `pam_auth_stack` and `SSHD_EFFECTIVE_COMMAND` are public (NetOrk/netork#662). ## [4.31.1] - 2026-11-06 ### Fixed - `get_log_sources()` names as a file's writer only a process that holds it open for writing. A log reader -- CrowdSec, Promtail, rsyslog's `imfile`, a `tail -F` -- was given the file it reads as if it wrote it. A file only readers hold is now listed without a producer (`unit` and `process` None); one only a reader inside a container holds is left out, as its path is that container's view of a mount. The access mode comes from `/proc//fdinfo/`, read in the same command (NetOrk/netork#2092). ## [4.31.0] - 2026-11-06 ### Added - `get_log_sources()` also says where the logs go beyond the host: `forwarding` holds whether rsyslog runs, its forwarding rules (legacy `@`/`@@`/`:omrelp:` and RainerScript `omfwd`/`omrelp`, with selector, ruleset and whether they take every message), its `imfile` inputs, and whether the journal reaches it (`imjournal`, or `imuxsock` with journald's `ForwardToSyslog`). Read in the same command; `parse_log_forwarding` and `parse_rsyslog` are public (NetOrk/netork#2053). ## [4.30.0] - 2026-11-04 ### Added - `LogSourcesMixin.get_log_sources()`: every log a Linux host writes. Sources are the units journald holds entries for, the log files running processes hold open (with size, last write and the service or container writing them), and the engine's file of each container's output. A file a container writes is named by the container's own path. `LOG_SOURCES_COMMAND` and `parse_log_sources` are public, and a host whose `find` cannot list open files (busybox) raises `LogSourcesUnavailable`. A driver supplies `_run_log_sources_command(command, privileged=)`, as for listening sockets (NetOrk/netork#2049). ## [4.29.0] - 2026-11-04 ### Added - `remove_local_account_keys(name, keys, *, clear_password=False)`: with `clear_password` the account's password goes along with the keys, and so do its sudo rule and admin groups, as sudo asks for that password. For a person who deleted their key and whose password went with it. The account stays open; the steps are `keys`, `password` and `sudo`, each run when another failed (NetOrk/netork#2033). ## [4.28.0] - 2026-11-04 ### Added - `LocalAccountsMixin.add_local_account(..., replaces=[...])`: the keys the new one takes the place of go from the account, matched by their base64 whatever a line's options or comment. Keys the person added on the host stay. The step is `old keys`; it runs even when the new key could not be added (NetOrk/netork#2013). - `add_local_account(..., clear_password=True)`: the account's password goes, and it logs in with a key only. Step `password`. It cannot be combined with a `password_hash` or with `admin=True`, as the sudo rule asks for that password. A driver supplies `_account_clear_password`; one that does not fails the step and says so. - `LocalAccountsMixin.remove_local_account_keys(name, keys)`: takes the named keys off an account and leaves it open, its password and sudo as they are -- for a person who deleted a key and has no new one. Step `keys`. - A driver whose host keeps keys elsewhere overrides `_account_drop_keys` next to `_account_add_key` and `_account_remove_keys`. ## [4.27.0] - 2026-11-03 ### Added - `ScheduledJobsMixin`: what a host runs on its own schedule as one shape, `ScheduledJobDict` — cron lines, run-parts scripts, anacron, systemd timers, `periodic`, a platform's own jobs, Windows tasks — with who owns each definition (`managed_by`) and what the driver can do to it (`actions`). `get_scheduled_jobs` reads them with the host's time zone; `manage_scheduled_job` enables, disables or runs one now; `put_scheduled_job` and `delete_scheduled_job` create, change and delete a job an application writes. Every write names the `digest` the caller saw (`job_digest`) and raises `ScheduledJobChanged` when the host's definition differs; `ScheduledJobRefused` for a job that cannot take a verb (NetOrk/netork#1933, #86). - `PosixScheduledJobsMixin`: the implementation over the command channel. One framed report as root reads `/etc/crontab`, `/etc/cron.d`, the user crontabs, the run-parts directories, `/etc/anacrontab` and every systemd timer with the unit it starts, and which package owns each definition. A cron line is switched off behind `#napalm-disabled: `; an application's job is a timer where systemd runs, else a file in `/etc/cron.d` or marked lines in the user's crontab. - `napalm_device_types.cron`: cron expressions as Vixie cron reads them, their next runs in a time zone, and as systemd `OnCalendar` lines that fire on exactly the same minutes. ### Removed - `OSDriver.get_cron_jobs` and `CronJobDict`. No application called them; the readers in napalm-linux and napalm-bsd go with them. `get_scheduled_jobs` replaces them. ## [4.26.0] - 2026-11-03 ### Added - `OpenSshServerMixin.SSHD_CONFIG_GENERATED`: a driver whose platform writes sshd_config from its own database, but includes `/etc/ssh/sshd_config.d` (OpenMediaVault), sets it and keeps `harden_ssh_server`. The hardening then goes only into the drop-in; a sshd_config without that include is left alone and the result says why, since the platform would write over a change to it on its next apply (#88, NetOrk/netork#1950). ## [4.25.0] - 2026-11-02 ### Added - `BackupServerDriver` (`ROLE = "backup_server"`), a role base for backup servers such as Proxmox Backup Server (#82). It declares `get_backup_datastores`, `get_backup_groups` (each group with its newest snapshot), `get_backup_snapshots`, `get_backup_jobs` (garbage collection, verify, prune, sync and tape in one list) and `get_backup_remotes`, and the actions `start_garbage_collection`, `run_backup_job`, `verify_backup_snapshot`, `set_backup_snapshot_protected` and `delete_backup_snapshot`, which return a task to follow with `get_backup_task`. The shapes live in `napalm_device_types.backups`. A server that manages its own disks lists `StorageDriver` as its second role. - `ClusterStorageDict.backup_server` (optional): which backup server, datastore and namespace a hypervisor's backup storage writes to, with the certificate fingerprint it pins, so an application can tie a server's backup groups to a cluster's guests. ## [4.24.0] - 2026-11-02 ### Added - `parse_lldpctl_keyvalue(output)` and `LLDPCTL_KEYVALUE_COMMAND`: lldpd's neighbours (`lldpctl -f keyvalue`) as one `LldpNeighborRow` per neighbour, for every driver of a host that runs lldpd. Several neighbours on one port (a phone and the PC behind it, an unmanaged switch) stay apart, an interface name with dots (`eth0.1`) stays whole, the chassis and port ids are read whatever their type (MAC, address, interface name, local id), with management addresses and capabilities. A failed lldpctl (lldpd not running, lldpctl missing) raises instead of reading as no neighbours. - `lldp_neighbors(rows)` and `lldp_neighbors_detail(rows, interface="")`: NAPALM's `get_lldp_neighbors` and `get_lldp_neighbors_detail` from those rows, so a driver no longer builds the two shapes itself. A neighbour that names its chassis by a MAC address carries it in `get_lldp_neighbors` as `mac`, in lower case (`LldpNeighborDict`), so an application can match it by MAC. ### Fixed - `DNF_SECURITY_COMMAND` reads the cached metadata only (`dnf --cacheonly`), as `APT_UPGRADABLE_COMMAND` reads apt's local index (#76). With an expired cache dnf downloaded the repositories' metadata first, and on an Oracle Linux host on OCI that outlasted netOrk's whole poll budget. Refreshing the index stays a step of its own (`dnf makecache`). Without a cache dnf fails at once, which already reads as "security unknown". ## [4.23.0] - 2026-11-02 ### Changed - Python 3.12 is the floor (`requires-python = ">=3.12"`): netOrk runs every driver on 3.12, so 3.10 and 3.11 are no longer tested (NAPALM/driver-standard#6). - The floor on napalm is 5.1: older releases import `pkg_resources`, which a Python 3.12 venv no longer has. - `requests` is a declared dependency (floor 2.25.0), for `tls_certificates`; it came with napalm before. - The SNMP health profile shapes (`SnmpHealthProfile`, `SnmpValueSpec`, `SnmpMemorySpec`, `SnmpInterfacesSpec`) are defined in `napalm_device_types.snmp_health_profile`. The package and `models` still export them, so an existing import keeps working. ### Added - `HypervisorDriver.get_vm_migration_preflight(name)`, `migrate_vm(name, target_node, online=True, target_storage=None, timeout=3600)` and `set_node_maintenance(node, enabled)`: which nodes of its cluster a guest could move to and what blocks it (`MigrationPreflightDict`: `allowed_nodes`, `refused_nodes` with the platform's reason, `local_disks` the migration copies, `local_resources` that block it), the move itself, blocking until the platform's task ended and raising with the platform's own cause, and a node's HA maintenance mode. Declared only; netOrk probes each with `hasattr`. - `get_clusters()` may report HA, backups and quorum votes: `ClusterDict.ha_resources` (`HAResourceDict`: requested and current state, node, PVE 8 group, PVE 9 rules), `backup_jobs` (`BackupJobDict`: schedule, storage, mode, the guests a job covers) and `backups` (`BackupDict`: per guest and storage the newest backup, its size and how many there are), and `ClusterNodeDict.votes`. All `NotRequired`, so drivers written before stay valid; None means could not be read, `[]` read and empty. - `napalm_device_types.tls_certificates`: TLS certificates by trust on first use, the counterpart of `ssh_host_keys` for a driver's HTTPS API. A driver takes the fingerprint to expect from `optional_args["tls_fingerprints"]` (by port, the SHA-256 of the DER certificate, `SHA256:AB:CD:…`) and refuses any other certificate with `TlsCertificateMismatch` (a `ConnectionException` naming both fingerprints) before a password is sent; no CA is consulted, so a self-signed certificate verifies. Without a fingerprint the first certificate is accepted. Every certificate seen goes to `optional_args["tls_certificate_observer"]`. `CertificateCheck.verify_server()` checks the certificate in a handshake of its own, `pin(session)` makes every later connection of a `requests` session present the same one, `refusing()` turns a refusal there into the mismatch, `observe()` reports it once the login stands (NAPALM/napalm-hpe-aruba-procurve#17). - `LldpAgentMixin.configure_lldp_agent(interfaces)` (`napalm_device_types.fixes`): run an LLDP agent on the interfaces the caller names. The template installs the agent (`ensure_package(LLDP_AGENT_PACKAGE)`), has the driver write the interface list (`_write_lldp_agent_config`), then enables and restarts the service (`manage_service(LLDP_AGENT_SERVICE, …)`); both names default to `lldpd`. No interface is a ValueError before anything changes. It is the action for what LLDP getters used to do on their own (NAPALM/napalm-openwrt#8). - `SnmpValueSpec.unknown`: a driver whose device's agent answers a value wrongly says so in its `SNMP_HEALTH_PROFILE` (`"cpu": {"unknown": True}`), and the application reports no value for it -- not the standard reading, not 0. OpenBSD's net-snmp, which reports the CPU as never idle, is the case (NAPALM/napalm-bsd#7). - `VMProvisionIncomplete` (`napalm_device_types.provisioning`): what `create_vm_from_cloud_init` raises when it fails after the VM was created. A `RuntimeError` with the failure as its cause and the half-built VM's `vmid` and `node`, so the caller can remove it with `destroy_vm` (NAPALM/napalm-proxmox#28). - `HypervisorDriver.get_clusters()` and its shapes `ClusterDict`, `ClusterNodeDict`, `ClusterStorageDict`, `ClusterNetworkDict`: every cluster the connection can see, with its nodes (offline ones included, `local` on the node the connection is), their CPU and memory, and the storages and networks with the nodes that see each. A standalone host is a cluster of one with no name; a manager reports each cluster and standalone host it runs. Declared only (NetOrk/netork#484, #1821). - `create_vm_from_cloud_init(node=…)`: the cluster node to create the VM on. None leaves it to the driver; a driver that can only create on its own node raises ValueError for another (NetOrk/netork#1822). - `LocalAccountsMixin.add_service_account(name, public_key)`: netOrk's own account, created through the account the driver logs in with today. It takes a key and no password, and a rule of its own in `85-service-` lets it run everything as root without a password (`NOPASSWD: ALL`, checked with `visudo` and `sudo -l`). An account that is there is opened again and keeps its other keys; root is refused (NetOrk/netork#1817). With `uid=`, a new account gets that uid once no other account has it; one that is there keeps its own. A driver supplies `_account_create_with_uid`. An account that is there and cannot log in (`nologin`, `false`) is another service's, such as a netOrk satellite's `netork`, and is refused before anything changes. - `LocalAccountsMixin` (`napalm_device_types.local_accounts`): personal login accounts on a host. `get_local_accounts(names)` reads them back (groups, lock, sudo, key files), `add_local_account(name, public_key, password_hash=, admin=)` creates or reopens one, `lock_local_account(name)` takes its keys and sudo away and refuses every login while keeping its home. Every command root runs is one fixed command, the key files are read and written as the account itself, and the sudo rule is a drop-in of its own, checked with `visudo` and `sudo -l`. A driver supplies the commands that create, open and lock an account and take it out of a group, and may replace where the keys are kept (`_account_key_files`, `_account_read_keys`, `_account_add_key`, `_account_remove_keys`; default `~/.ssh/authorized_keys`) (NetOrk/netork#1793). - `OpenSshServerMixin.exempt_from_ssh_penalties`: puts the address the host sees the managing client at (`$SSH_CONNECTION`) into `PerSourcePenaltyExemptList`, in `/etc/ssh/sshd_config.d/00-netork-penalty-exempt.conf`, next to the host's own exemptions. Only where sshd has per-source penalties (OpenSSH 9.8+; an older one refuses the directive) and sshd_config includes the drop-in directory; checked with `sshd -t` and `sshd -T`, then a new session, and taken back on any failure. Helpers: `parse_penalty_exemptions`, `client_address`, `is_exempt` (NetOrk/netork#1800). - `retry_refused` and `refused_before_banner` (`napalm_device_types.ssh_host_keys`): a connection sshd closes before its banner -- what OpenSSH 9.8+ does to a source it has just penalised -- is tried again after 1, 3, 7 and 15 s. `HostKeyCheck.connect` does so itself (NetOrk/netork#1800). - `ARP_AGE_UNKNOWN` (`napalm_device_types.arp`, `-1.0`): the `age` a driver reports in `get_arp_table` and `get_ipv6_neighbors_table` when the device does not say how old an entry is. `0.0` means "confirmed just now" and is no stand-in for it (NetOrk/netork#1327). - `NetworkTargetDict.uplinks` (optional): the physical NICs a bridge leaves the host through, one group per way out -- a bond's slaves together, a single NIC alone (`[["eno1", "eno2"]]`). A missing key means unknown, an empty list a bridge without uplink. NetOrk follows their LLDP neighbours to the switch port and its untagged VLAN; one slave's neighbour speaks for its bond (NetOrk/netork#1325). - `HostShutdownMixin.shutdown_host()`, on `DeviceTypeDriver`: shut the device itself down cleanly through its own means; declared, so `hasattr` says whether a driver can. No forced variant (NetOrk/netork#1636). - `VmStopRequestMixin.begin_stop_vm(name)`: ask for a VM's clean shutdown and return at once, for callers that wait on a timer instead of in `stop_vm`. No role demands it (NetOrk/netork#1636). - `SudoHistoryMixin` (`napalm_device_types.sudo_history`): the sudo password an earlier driver typed into an interactive shell (`echo '' | sudo -S -p "" …`) and left in the login user's shell history, for a driver to mix in (NetOrk/netork#1090). The driver says what its lines carry (`SUDO_HISTORY_MARKER`) and which files its partial sudo left to the login user (`SUDO_ROOT_FILES`), and supplies `run_command`. - `get_history_sudo_lines()` counts the lines per history file (`~/.bash_history`, `~/.zsh_history`, `~/.ash_history`, `~/.sh_history`) with `grep -c` as the login user. No line is read back; a file it cannot read raises. - `remove_history_sudo_lines()` deletes exactly those lines as the login user (`sed -i`, its backup overwritten with `shred -u`), and as root gives each of `SUDO_ROOT_FILES` back to `root:root` where it is a regular file, not a symbolic link, owned by the login user. It reports every file's lines before and after, how many were removed and which files are root's again. Nothing goes to stdin. ### Fixed - `AptProxyMixin` knows that apt's https sources take the http proxy too: apt's https method asks `Acquire::https::Proxy` and then `Acquire::http::Proxy`, for the host and then for all. An `00proxy` with the http line alone sent download.docker.com and every other https source to apt-cacher-ng, which answers `403 CONNECT denied` without a `PassThroughPattern`, and the whole upgrade failed (NetOrk/netork#1811). - The report carries `https_proxy`, `https_per_host` and `https_settings`; `https_proxy_for(report, host)` is the proxy apt takes for one https host. - `check_https_source()` fetches one https source's `InRelease` the way apt does; None where no https source goes through a proxy. - `configure_package_proxy` gains an `https` step: one https source through the new proxy. Where the proxy refuses it, `00proxy` also says `Acquire::https::Proxy "DIRECT";`, later https lines to the same proxy are commented out, and the source is fetched direct. An HTTP error from the mirror counts as let through. - `apt_fetch_command` takes https URLs; `direct` sets the host's https setting too. - `comment_out_command(..., scheme="https")`. ## [4.22.0] - 2026-10-23 ### Added - `AptProxyMixin` (`napalm_device_types.apt_proxy`): apt's proxy on a Debian-family host, for a driver to mix in (NetOrk/netork#1397). - `get_package_proxy()` is the proxy apt really fetches through, as `apt-config dump` says, and no longer the content of one file. A proxy the Debian or Ubuntu installer set elsewhere wins over `00proxy`, and was missed. - `get_package_proxy_report()` adds where it is set: every line of apt's configuration that sets it, in the order apt reads them, the per-host proxies, and the sources fetched over http and over https. - `check_package_proxy(url, expect)` fetches a page from the host with apt's own downloader, past any proxy, and says whether the host reached it and whether the page says what you expected. - `configure_package_proxy(url)` writes the proxy, comments out every other line that sets a different one, reads the result back and fetches one source through the proxy. It fails when apt still fetches elsewhere. - `same_proxy_url()` compares two proxy URLs by scheme, host and port. ## [4.21.1] - 2026-10-22 The first release on PyPI since 0.1.0. ### Changed - The package follows the NAPALM driver standard (https://git.netork.io/NAPALM/driver-standard): CI on Gitea runs ruff, strict mypy, the tests on Python 3.10–3.13 and at the lowest declared versions, bandit, pip-audit and a build check, and a release tag publishes to PyPI. - Type annotations use builtin generics (`list[str]`, `X | None`) instead of `typing.List` and `Optional`. They mean the same to a type checker. - The project links point at git.netork.io instead of GitHub.